From: Danny Xu (quantum_mania_at_yahoo.com)
Date: Fri Jun 10 2011 - 22:34:34 CDT
Jim,
Yes, I got the full message directly from Ajas. See below.
Looks like GPU only accelerates large systems, over 100k atoms.
DX
________________________________
From: Ajasja Ljubetič <ajasja.ljubetic_at_gmail.com>
To: Danny Xu <quantum_mania_at_yahoo.com>
Cc: Norman Geist <norman.geist_at_uni-greifswald.de>; Francesco Pietra
<chiendarret_at_gmail.com>; Namd Mailing List <namd-l_at_ks.uiuc.edu>
Sent: Fri, June 10, 2011 12:17:43 PM
Subject: Re: namd-l: max number of CPU PEs per GPU?
Hi,
Sometimes it seems like the mailing list sends only half my message... (I have
not yet had time to look for the problem). I hope this time the whole mail gets
through.
I
just finished benchmarking a GTX 470 on a tyan mother board with two AMD
Opteron Quad 2376.
Here are the results (red is with GPU, black without). One can see that the
optimal CPU:GPU ratio is very size dependant, but in general more than 8 cores
is never a good idea.
Best regards,
Ajasja
On Fri, Jun 10, 2011 at 18:20, Danny Xu <quantum_mania_at_yahoo.com> wrote:
What are your benchmark findings?
>
>
>
________________________________
From: Ajasja Ljubetič <ajasja.ljubetic_at_gmail.com>
>To: Norman Geist <norman.geist_at_uni-greifswald.de>
>Cc: Francesco Pietra <chiendarret_at_gmail.com>; Namd Mailing List
><namd-l_at_ks.uiuc.edu>
>Sent: Fri, June 10, 2011 3:50:44 AM
>Subject: Re: namd-l: max number of CPU PEs per GPU?
>
>
>Hi,
>
>I just finished benchmarking a GTX 470 on a tyan mother
>board<http://www.tyan.com/product_board_detail.aspx?pid=453> with
>two AMD Opteron Quad
>2376<http://products.amd.com/en-us/OpteronCPUDetail.aspx?id=489>
This archive was generated by hypermail 2.1.6 : Mon Dec 31 2012 - 23:20:25 CST